1/5 As part of our work on improving the efficiency of our LLM online-RL training pipelines, we cut policy update step time by ~70% by introducing a model-agnostic padding minimization method.
LLMS
-
DeepSeek’s Architectural Legacy Dominates Chinese AI Development
By
–
They’re not gone. Most Chinese AI companies are still building on architectures derived from DeepSeek.
If you actually read the research papers, it’s obvious. Hype cycles move fast.
Architectural influence lasts much longer. -
ChatGPT auto-installing Leboncoin app: hidden feature and security risk
By
–
Example here, I would like ChatGPT to automatically install the Leboncoin app. If they do that, ok, the ROI will be positive. But in the meantime, it remains a hidden feature which, I think, will be used almost not at all, plus the security risk.
-
Hacking Leboncoin with ChatGPT in 5 seconds
By
–
Can anyone hack Leboncoin with ChatGPT now? I wanted to test it out and well… Leboncoin spent years protecting itself from scrapers and data dumps… only for me to ultimately extract their data in 5 seconds via ChatGPT. I love it (or not at all…). In 5 seconds, I can
-
Multimodal AI model gains text-to-video and image-to-video
By
–
2. Upgraded audio. You can now pinpoint the exact character speaking.
— Robert Scoble (@Scobleizer) 11 février 2026
3. Much improved text. Precise lettering capabilities. Great for making visuals for slide decks and web sites. And, of course, X posts.
4. Unified model. You can do text-to-video. Image-to-video.… pic.twitter.com/tYde4uzBt42. Upgraded audio. You can now pinpoint the exact character speaking. 3. Much improved text. Precise lettering capabilities. Great for making visuals for slide decks and web sites. And, of course, X posts. 4. Unified model. You can do text-to-video. Image-to-video.
-
Human-in-the-Loop Review Workflows for LLM Applications and Agents
By
–
Human-in-the-Loop Review Workflows for LLM Applications & Agents https://
buff.ly/Pt9EanI
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Claude Opus 4.6 Sabotage Risk Report Released
By
–
When we released Claude Opus 4.5, we knew future models would be close to our AI Safety Level 4 threshold for autonomous AI R&D. We therefore committed to writing sabotage risk reports for future frontier models. Today we’re delivering on that commitment for Claude Opus 4.6.
-
Anthropic releases Opus 4.6 AI safety risk assessment report
By
–
Rather than making difficult calls about blurry thresholds, we decided to preemptively meet the higher ASL-4 safety bar by developing the report, which assesses Opus 4.6’s AI R&D risks in greater detail. Read the sabotage risk report here: https://
anthropic.com/claude-opus-4-
6-risk-report
…
